Classifications

• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T7/00—Image analysis
  • G06T7/0002—Inspection of images, e.g. flaw detection
  • G06T7/0004—Industrial image inspection
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2207/00—Indexing scheme for image analysis or image enhancement
  • G06T2207/10—Image acquisition modality
  • G06T2207/10116—X-ray image
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2207/00—Indexing scheme for image analysis or image enhancement
  • G06T2207/10—Image acquisition modality
  • G06T2207/10116—X-ray image
  • G06T2207/10121—Fluoroscopy
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2207/00—Indexing scheme for image analysis or image enhancement
  • G06T2207/30—Subject of image; Context of image processing
  • G06T2207/30108—Industrial image inspection
  • G06T2207/30148—Semiconductor; IC; Wafer

Definitions

• This disclosure generally relates to techniques for monitoring semiconductor processes during production by non-destructively measuring layer thickness and composition on structures using photoelectron spectroscopy and x-ray fluorescence.
• Integrated circuits typically comprise a number of layers formed on a silicon substrate. As integrated circuits become smaller, and the thickness of the layers comprising the integrated circuits is reduced, the behavior of devices formed from these layers often depends on the thickness or composition of a specific layer. For example, a transistor formed on a silicon substrate may have different characteristics depending on the thickness or composition of the gate of the transistor. Therefore, during fabrication it is highly beneficial to monitor the processes by determining the thickness and composition of layers in the fabricated microelectronic device such as an integrated circuit.
• XPS x-ray photoelectron spectroscopy
• XRF x-ray fluorescence spectroscopy
• Various disclosed embodiments provide methods and systems for improved signal acquisition in XPS and XRF systems.
• the disclosed embodiments are especially suitable for monitoring the consistency of the processes during the production of ICs.
• the embodiments enable investigating characteristics such as composition and thickness of thin films layered over a substrate.
• Disclosed embodiments have been demonstrated to provide improved analysis results while reducing the standard signal acquisition time by half.
• an XPS or an XRF tool is used to collect emissions from the wafer in a conventional manner, except that the acquisition time is reduced, e.g., by 50%.
• the collected dataset is rather noisy, having relatively low signal to noise ratio.
• the obtained dataset is therefore unsuitable for standard analysis of process monitoring.
• the quality of the dataset is therefore improved by operation of an unsupervised machine learning.
• An important aspect of the unsupervised machine learning is its ability to segregate variability in the dataset that contribute to noise, but do not contribute to the ability to identify excursions in the process. The unsupervised machine learning is therefore used to reduce variability in the spectra due to noise.
• Disclosed embodiments provide methods for monitoring process excursions in semiconductor processing of an integrated circuit (IC), comprising the steps of irradiating the IC to thereby generate emissions from the IC; collecting the emissions from the IC using one of x- ray photoelectron spectroscopy (XPS) or x-ray fluorescence spectroscopy (XRF) and generating from the emissions a dataset corresponding to count per unit time versus kinetic energy; performing principal component analysis on the dataset to thereby obtain principal component values from the dataset; selecting a number of principal components that exhibit high variance contribution compared to remaining principal components; removing from the dataset all values corresponding to the remaining principal components to thereby obtain filtered dataset; and analyzing the filtered dataset to determine presence of process excursion.
• XPS x- ray photoelectron spectroscopy
• XRF x-ray fluorescence spectroscopy
• a method for enhanced statistical process control comprising the steps: obtaining dataset corresponding to photoelectron emission from a sample, the dataset representing the spectra of the photoelectrons emitted from the sample; applying principal component analysis to the dataset to obtain principal component variance for each principal component; examining the variances to select an N number of relevant principal components having the highest variance values; selecting from the dataset all relevant data points belonging to the N number of relevant principal components and using only these data points to update the spectra; using the updated spectra to calculate photoelectron emission intensity; plotting the calculated photoelectron emission intensity on a statistical process control (SPC) chart; and inspecting the SPC chart to identify process excursions.
• SPC statistical process control
• a metrology module for monitoring fabrication process of an integrated circuit (IC).
• the module comprises: an input module receiving signal indicative of emissions from the IC, e.g., from an XPS or XRF metrology tool; a principal component analysis (PCA) module receiving the signal and calculating therefrom variance corresponding to each principal component; a PCA number selector module selecting a number of N principal components according to the calculated variance; a filtering module that selects from the original XPS signal only the data points corresponding to the selected N principal components; an intensity module that uses the selected data points to calculate an intensity value of the emissions; a conversion module converting the emission intensity into material parameter; and a display module that displays the calculated parameter on a monitor, generally in the form of an entry in a PCA chart.
• PCA principal component analysis
• FIG. 1 Other aspects provide a machine readable medium having stored thereon executable program which, when executed, causes a machine to perform a method for monitoring a semiconductor fabrication process, the method comprising: receiving a dataset corresponding to photon count per unit time versus kinetic energy of emitted photons obtained by collecting the emissions from the IC using x-ray photoelectron spectroscopy (XPS) and generating the dataset from the emissions; performing principal component analysis on the dataset to thereby obtain principal component values from the dataset; selecting a number of principal components that exhibit high variance contribution compared to remaining principal components; removing from the dataset all values corresponding to the remaining principal components to thereby obtain filtered dataset; and analyzing the filtered dataset to determine presence of process excursion.
• XPS x-ray photoelectron spectroscopy
• FIG. l is a plot of dataset obtained using XPS measurement, according to disclosed embodiment
• FIG. 2 illustrates a plot of principal component analysis (PCA) performed on the dataset of Fig. 1, according to one embodiment
• FIG. 3 is a logarithmic plot of the principal component variance versus the principal component number for the plot of Fig. 2, according to an embodiment
• FIG. 4 illustrate a plot of filtered dataset obtained by applying PCA to the XPS measurement, according to disclosed embodiment
• FIG. 5A illustrates a plot of data obtained using full time acquisition versus data acquired using half time acquisition
• Fig. 5B illustrates a plot of data obtained using full time acquisition versus interpolated data, which is data calculated by interpolating half of the full-time data points, according to disclosed embodiment
• FIG. 6 A illustrates the within wafer standard deviation of boron percentage calculated from dataset taken at full time (filled dots) and half time (circles), while Fig. 6B illustrates the standard deviation of boron percentage calculated from dataset taken at full time (filled dots) and a filtered half-time dataset (open circles), i.e., after the half time dataset has been reduced to remove data corresponding to principal components not selected within the N selected principal components, according to disclosed embodiment;
• FIG. 7 illustrate an SPC chart of three different XPS tools wherein the black dots represent within-wafer averaged boron percentage calculated from data taken in full time acquisition, while the circles represent within-wafer averaged boron percentage calculated from enhanced half-time data using the embodiments described herein;
• FIG. 8 is a flow chart for a method of performing enhanced statistical process control, according to an embodiment
• FIG. 9 is a block diagram illustrating a metrology module for monitoring fabrication process of an integrated circuit, according to an embodiment.
• Embodiments disclosed herein improves the operation and efficiency of XRF and/or XPS process monitors.
• the reader is fully familiar with the construction and operation of such monitors, which can be gleaned from e.g., Applicant’s U.S. patents 10,533,961 and 10,801,978, the disclosures of which are incorporated herein in their entirety for completeness.
• the “noisier” the XPS or XRF signal obtained during monitoring becomes. This leads to increase in the signal acquisition time so as to increase the signal to noise ratio.
• an important parameter in SiGe:B fabrication of raised source and drain (RSD) transistors is the percentage of boron in the resulting layer.
• Another important parameter is the thickness of a TiN layer, e.g., in gate electrodes.
• Embodiments disclosed herein have been demonstrated to result in halving the acquisition time without any reduction in accuracy for measuring boron percentage in SiGe:B structures.
• the disclosed embodiments were demonstrated to improve throughput capability by 70% for measuring TiN film thickness. Conversely, when utilizing the same acquisition time, the embodiments were demonstrated to improve precision by 30%.
• Fig. 1 is a plot of dataset obtained using XPS measurement of boron concentration in wafers having SiGe:B structures formed therein, showing counts per unit time (here, counts per second) versus bin numbers, wherein each bin corresponds to certain binding energy level (normally in eV units).
• This data can be used for machine training and may be obtained, e.g., by a standard XPS acquisition (e.g., 1.5-3 minutes of acquisition time) on 13 sites over the wafer, taking 20 measurements at each site.
• the data is not readily decipherable to determine whether the resulting product meets the specification, or whether an excursion occurred in the process leading to changes in the boron percentage.
• unsupervised machine learning process has been implemented to remove data points that do not contribute to the analysis of whether excursions occur in the fabrication process.
• unsupervised machine learning approach is unique in this context, as the machine is not provided with labeled or scored reference data. For example, in the context of boron concentration, the machine is not provided with reference data indicating the parameters that are characteristic of the proper boron concentration. Instead, the machine needs to self-discover any naturally occurring patterns in the dataset and determine which parameters are important to identifying deviation from the desired process results, e.g., deviation from the desired boron concentration.
• unsupervised machine learning is implemented using principal component analysis (PCA).
• PCA principal component analysis
• Figure 2 illustrates a plot of PCA performed on the dataset of Fig. 1, showing the principal component value versus the measurement number.
• Figure 3 is a logarithmic plot of the principal component variance versus the principal component number. As can be observed by the dashed oval, there are N (here four) principal components that contribute significantly to the variance, while the remaining principal components contain data which does not materially contribute to the variance.
• the number of principal components called a hyperparameter, may be preset for the machine learning or may be part of the selection algorithm to be set automatically during the PCA process.
• a reverse PCA process is applied to return to the energy space.
• the dataset plot shown in Fig. 1 is recreated, but using only data belonging to the N principal components that contribute meaningfully to the variance.
• the plot of Fig. 4 is much “cleaner” than that of Fig. 1, although it corresponds to the same dataset.
• the data of Fig. 4 can now be analyzed to determine the quality of the SiGe:B deposition process.
• One consequence of the disclosed embodiment is that the machine learning does not rely on long training period and does not require repetition of measurements to be used as reference. Rather, by performing the outlined steps, the data of the selected N principal components becomes the reference itself. Any standard algorithm for determining composition or film thickness can be directly applied to the reconstructed dataset, which includes only data points from the selected N principal components. In fact, it has been demonstrated that this approach results in improved accuracy compared to applying the standard algorithm to the initial dataset. Conversely, it has been shown that the acquisition time can be reduced and by employing the disclosed process the same accuracy can be achieved as with standard acquisition time. Acquisition time may be reduced by reducing the time of signal acquisition or by taking fewer measurements at each site.
• Figure 5A illustrates a plot of data obtained using full time acquisition versus data acquired using half time acquisition (in this case, using half the number of repetitions per site).
• Figure 5B illustrates a plot of data obtained using full time acquisition versus interpolated data, which is data acquired using full-time interpolated acquisition with half the number of points from the spectra.
• Fig. 6A illustrates the standard deviation of boron percentage calculated using dataset taken at full time (filled dots) and half time (open circles).
• Figure 6B illustrates the standard deviation of boron percentage calculated using dataset taken at full time (filled dots) and filtered half time (open circles), i.e., after the half time dataset has been reduced to remove data corresponding to principal components not selected within the N selected principal components.
• the fit in Fig. 6B is much tighter after the PCA process is applied to the half time dataset shown in Fig. 6A.
• the process proceeds by obtaining XPS spectra from a wafer undergoing inspection.
• the unsupervised machine learning process then applies PCA to the spectra data.
• a number of N principal components that generate the highest variation are selected and the data corresponding to these N principal components are used to regenerate the spectra curve.
• the process then proceeds to determine the intensity by calculating the area under the curve.
• the intensity is then converted to the desired value (e.g., boron concentration, TiN thickness, etc.) using a predetermine relationship, i.e., the intensity value correlates to the inspected material property. For example, different boron concentration in the layer would result in different emission intensity.
• the intensities or the corresponding material property values can be plotted in an SPC (statistical process control) chart in order to monitor the fabrication process.
• Fig.7 illustrate an SPC chart of boron percentage obtained from three different XPS tools, wherein the black dots represent boron concentration calculated using data taken in full time acquisition, while the circles represent boron concentration calculated using enhanced half-time data using the embodiments described herein. As can be seen there is a good agreement of the data, showing the process is within the allowable variance.
• Fig. 8 starts at step 800 by obtaining dataset corresponding to photoelectron emission from a sample.
• the dataset represents the spectra of the photoelectrons emitted from the sample.
• the process proceeds by applying principal component analysis to the dataset to obtain principal component variance for each principal component.
• the variances are examined to select a number of N relevant principal components having the highest variance values. This can be done by plotting all the variances in descending or ascending order and detecting an inflection point. All variances having values higher than the inflection points are considered to belong to the relevant principal components and are selected as part of the group of N principal components.
• the process proceeds by reverting to the energy space. This is done by selecting from the dataset all relevant data points belonging to the N number of relevant principal components and using only these data points to replot the spectra.
• the updated replot spectra is used to calculate photoelectron emission intensity, i.e., the calculation is performed using only the relevant data points.
• the calculated photoelectron emission intensity is converted to layer property by a known relationship. For example, boron percentage or TiN thickness correlates with emission intensity and this known relationship is used to convert intensity to material property.
• the material property is plotted the on a statistical process control (SPC) chart. Finally, at step 835 the SPC chart is inspected to identify process excursions.
• SPC statistical process control
• FIG. 9 is a block diagram illustrating a metrology module for monitoring fabrication process of an integrated circuit (IC).
• the module includes an input module 900 receiving signal indicative of photoelectrons emitted from the IC, e.g., from an XPS metrology tool.
• the entire metrology module may be integrated in an XPS or XRF metrology tool, e.g., implemented as software, hardware, or a combination of both, or may be implemented as part of the controller of the metrology tool, such as a general purpose computer or a specifically tailored computer.
• the input module 900 delivers the signal to a principal component analysis (PCA) module 905.
• the PCA module 905 receives the signal and calculates therefrom variance corresponding to each principal component.
• a PCA number selector module 910 selects a number of N principal components according to the calculated variance.
• a filtering module 915 selects from the original XPS signal only the data points corresponding to the selected N principal components.
• An intensity module 920 uses the selected data points to calculate an intensity value of the photoelectron emissions, i.e., calculates the area under the curve of a plot of the selected data points.
• the conversion module 925 applies a known relationship to convert the calculated intensity to a material property value (e.g. concentration percentage or layer thickness).
• the display module 930 displays the material property value on a monitor, generally in the form of an entry in a PCA chart.
• a method for monitoring process excursions in semiconductor processing of an integrated circuit comprising the steps of: irradiating the IC to thereby generate emissions from the IC; collecting the emissions from the IC using one of x-ray photoelectron spectroscopy (XPS) or x-ray fluorescence spectroscopy (XRF) and generating from the emissions a dataset corresponding to photon count per unit time versus kinetic energy of emitted photons; performing principal component analysis on the dataset to thereby obtain principal component values from the dataset; selecting a number of N principal components that exhibit high variance contribution compared to remaining principal components; removing from the dataset all values corresponding to the remaining principal components to thereby obtain filtered dataset; analyzing the filtered dataset to determine presence of process excursion.
• XPS x-ray photoelectron spectroscopy
• XRF x-ray fluorescence spectroscopy
• the number N may be preset or selecting a number of N principal components may comprise calculating variance for each principal component and plotting the variance versus corresponding principal component to identify an inflection point; and setting N to correspond to the number of principal components having higher variance from the inflection point.
• Analyzing the filtered dataset may comprise calculating emission intensity from the filtered dataset, converting the emission intensity to material property value, and plotting the material property value on a statistical process control chart.
• a metrology module for monitoring fabrication process of an integrated circuit comprising: an input module receiving signal indicative of photoelectrons emitted from the IC; a principal component analysis (PCA) module receiving the signal and calculating therefrom variance corresponding to each principal component; a PCA number selector selecting a number of N principal components according to the calculated variance; a filtering module selecting from the signal data points corresponding to the N principal components; an intensity module calculating emission intensity from the output of the filter module; a conversion module converting the emission intensity into material property value; and a display module plotting the material property value.
• PCA principal component analysis
• the input port may be coupled to a sensor of an x-ray photoelectron spectroscopy (XPS) or to a sensor of an x-ray fluorescence spectroscopy (XRF).
• XPS x-ray photoelectron spectroscopy
• XRF x-ray fluorescence spectroscopy
• a machine readable medium having stored thereon executable program code which, when executed, causes a machine to perform a method for determining anomalies in a property of a fabricated layer of a sample, the method comprising: collecting photoelectron emissions from the sample using one of x-ray photoelectron spectroscopy (XPS) or x-ray fluorescence spectroscopy (XRF) and generating from the emissions a dataset corresponding to photoelectron count per unit time versus kinetic energy of emitted photoelectrons; performing principal component analysis on the dataset to thereby obtain principal component values from the dataset; selecting a number of principal components that exhibit high variance contribution compared to remaining principal components; removing from the dataset all values corresponding to the remaining principal components to thereby obtain filtered dataset; and analyzing the filtered dataset to determine presence of process excursion. Analyzing the filtered dataset may include calculating total intensity of photoelectron emission using the filtered dataset, converting the intensity into a material property value, and plotting the material property

Landscapes

• Engineering & Computer Science (AREA)
• Quality & Reliability (AREA)
• Computer Vision & Pattern Recognition (AREA)
• Physics & Mathematics (AREA)
• General Physics & Mathematics (AREA)
• Theoretical Computer Science (AREA)
• Analysing Materials By The Use Of Radiation (AREA)
• Investigating, Analyzing Materials By Fluorescence Or Luminescence (AREA)

Applications Claiming Priority (2)

Publications (1)

Family

ID=87000290

Family Applications (1)

Country Status (2)

Citations (2)

• 2022
  • 2022-12-27 TW TW111150168A patent/TW202343611A/zh unknown
  • 2022-12-30 WO PCT/US2022/054302 patent/WO2023129690A1/en unknown

Patent Citations (2)

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events